In the years around 2015, a broad dissatisfaction with “equilibrium-first” economics—sharpened by the 2008 global financial crisis and the uneven recovery that followed—helped propel complexity economics from a research frontier into a more visible, policy-adjacent conversation. The problem was not that equilibrium models were useless, but that many of the era’s most damaging phenomena looked like quintessential complex-system events: cascades of leverage and margin calls, network contagion, liquidity evaporation, and feedback between beliefs and prices. These dynamics were hard to reconcile with representative-agent frameworks and small, mean-reverting deviations around a stable steady state. Complexity economics, associated with researchers such as W. Brian Arthur (Santa Fe Institute) and Doyne Farmer (Oxford’s Mathematical Institute and the Institute for New Economic Thinking), offered a different starting point: the economy as an evolving system of heterogeneous, interacting agents that may never settle into equilibrium.
Historical context: what problems were being addressed?
Complexity economics drew on intellectual currents that predate 2015 by decades: nonlinear dynamics, statistical physics, computer science, and evolutionary thinking. Arthur’s early work on “increasing returns” and path dependence in the 1980s and 1990s challenged the standard presumption of diminishing returns and unique equilibria. His influential paper “Competing Technologies, Increasing Returns, and Lock-In by Historical Events” (Economic Journal, 1989) formalized how small historical accidents can select a dominant technology—an outcome difficult to square with tidy equilibrium selection stories. By the 2000s, agent-based modeling (ABM) had matured computationally, while finance—where Farmer built models of market ecology and microstructure—offered plentiful high-frequency data displaying heavy tails, volatility clustering, and extreme events. Around 2015, these strands converged with a policy demand: stress testing, macroprudential regulation, and systemic-risk monitoring required tools that could represent networks, heterogeneity, and nonlinear feedback.
Core ideas and key concepts
Complexity economics replaces the equilibrium metaphor with a process metaphor. The economy is an adaptive system: agents learn, revise strategies, and co-evolve with institutions and technologies. Several concepts are central:
-
Heterogeneous agents and bounded rationality. Instead of a representative agent optimizing with full information, agents use heuristics, limited foresight, and learning rules. Aggregate outcomes emerge from interactions, not from a single optimizing problem.
-
Non-equilibrium dynamics. Rather than assuming markets clear continuously, complexity models allow persistent out-of-equilibrium states: unemployment, excess capacity, bubbles, and crises can be endogenous and long-lived.
-
Feedback, increasing returns, and path dependence. Positive feedback—via network effects, learning-by-doing, or expectations—can generate multiple possible outcomes. Arthur summarized this intuition in Increasing Returns and Path Dependence in the Economy (1994): “Small events are not averaged away and ‘forgotten’ … they may decide the outcome.”
-
Emergence and self-organization. Macro “laws” (like business cycles or firm size distributions) can emerge from micro rules, akin to how flocking emerges from simple behavioral rules in biology.
-
Networks and systemic risk. Links among banks, firms, and households transmit shocks. Contagion depends on topology (concentration, centrality), leverage, and behavioral responses.
Farmer’s work in finance helped popularize a related claim: market outcomes reflect an ecology of strategies under constraints, not instant equilibration. His empirical emphasis—fat-tailed returns and clustered volatility—pushed complexity economists to calibrate models to data, rather than treating stylized facts as anomalies.
Mathematical formulation: from equilibrium to interacting stochastic processes
Complexity economics is not one model but a toolkit. Still, many ABMs share a common mathematical skeleton: a high-dimensional stochastic dynamical system.
Let (i=1,\dots,N) index agents with state variables (x_i(t)) (wealth, inventory, beliefs, leverage). Let (p(t)) denote prices and (s(t)) exogenous or slow-moving states (technology, policy). Agents update via behavioral rules: $$ x_i(t+1)=F_i!\big(x_i(t),\, p(t),\, \mathcal{I}i(t),\, \eta_i(t)\big), $$ where (\mathcal{I}_i(t)) is agent (i)’s information set and (\eta_i(t)) is idiosyncratic noise. Prices are determined by an aggregation/market-clearing mechanism rather than an equilibrium condition: $$ p(t+1)=G!\left(p(t),\, \sum{i=1}^N q_i(t),\, \xi(t)\right), $$ with (q_i(t)) representing demands/orders generated from (x_i(t)) and (\xi(t)) capturing market-level shocks or microstructure noise. Importantly, “clearing” may be partial, delayed, or mediated by inventories and credit constraints.
A common learning specification uses reinforcement or evolutionary selection. If each agent chooses among strategies (k\in{1,\dots,K}) with propensities (w_{ik}(t)), one may write: $$ w_{ik}(t+1)=w_{ik}(t)\exp!\big(\beta \pi_{ik}(t)\big), \qquad P_{ik}(t)=\frac{w_{ik}(t)}{\sum_{j=1}^K w_{ij}(t)}, $$ where (\pi_{ik}(t)) is the realized payoff and (\beta) is intensity of selection. In the large-population limit, this resembles replicator dynamics: $$ \dot{s}_k = s_k\big(\pi_k - \bar{\pi}\big), $$ a formal bridge to evolutionary game theory.
Where equilibrium economics often seeks closed-form comparative statics, complexity economics leans on simulation and econometric validation: matching distributions (e.g., firm sizes), moments (volatility clustering), and scenario-based stress outcomes. Around 2015, advances in computation and estimation—indirect inference, simulated method of moments, and Bayesian calibration for ABMs—made it increasingly feasible to confront ABMs with data, though identifiability remains challenging.
Historical impact and criticism
By 2015, complexity economics had influenced several domains: systemic-risk research, network-based stress testing, market microstructure models, and innovation/technology diffusion analysis (where Arthur’s increasing-returns logic is especially natural). It also helped legitimize pluralism in macro modeling—alongside DSGE, a growing ecosystem of HANK models, network models, and ABMs.
Critics raise three recurring objections. First is opacity: large ABMs can become “black boxes,” where results hinge on poorly understood mechanisms. Second is calibration and identification: many parameter sets can fit the same stylized facts, complicating inference and policy use. Third is normative ambiguity: without a welfare theorem or a crisp social planner benchmark, evaluating policy can be less straightforward. Complexity economists respond that real economies may not satisfy the conditions under which clean welfare results hold, so realism requires new evaluative tools (robustness checks, counterfactual simulation, and stress testing rather than point predictions).
Legacy and relevance today
The legacy of complexity economics is a reframing of what counts as a “macro foundation.” Instead of deriving macro from a single optimization problem, it derives macro from interaction, institutions, and adaptive behavior. Its relevance has only grown: platform economies and AI-driven markets feature network effects and increasing returns; climate transition risks propagate through financial and supply-chain networks; and episodes like the 2020 liquidity scramble and subsequent inflation shocks underscore nonlinear dynamics and regime shifts. Complexity models are also aligned with modern policy concerns: resilience, fragility, and tail risks, not just average outcomes.
Line of continuity: from classical equilibrium to complex adaptive systems
Complexity economics is best seen as an extension and critique of the Walrasian and Marshallian equilibrium tradition, formalized in Arrow-Debreu general equilibrium (1954) and later embedded in rational expectations macroeconomics. It shares with earlier traditions—Schumpeter’s emphasis on innovation and “creative destruction” (1939, 1942), Keynes’s focus on uncertainty and shifting expectations (1936), and Hayek’s view of markets as information-processing systems (1945)—a skepticism that equilibrium is the natural resting place. Arthur’s increasing-returns framework connects directly to Schumpeterian dynamics and industrial organization, while Farmer’s empirically disciplined finance work connects to econophysics and to post-crisis calls for models that take instability seriously.
After 2015, complexity economics has increasingly interacted with neighboring approaches rather than positioning itself purely as an alternative: network macro, behavioral economics, and empirical microstructure have all cross-pollinated with ABMs. The continuing project is not to abolish equilibrium theory, but to put it in its place—as one possible regime within a broader, non-equilibrium, evolving economic landscape.